whisper-large-v3-kor_eng_large-v3_ed_ev

This model is a fine-tuned version of openai/whisper-large-v3 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0001
• Cer: 2486.3636

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 4
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss
4.4137	2.2775	100	2515.9091	1.4492
0.7675	4.5549	200	254.5455	0.4655
0.3364	6.8324	300	147.7273	0.2105
0.1142	9.2543	400	143.1818	0.0351
0.0122	11.5318	500	97.7273	0.0032
0.0021	13.8092	600	111.3636	0.0014
0.0011	16.0694	700	111.3636	0.0009
0.0008	18.3468	800	377.2727	0.0006
0.0006	20.6243	900	852.2727	0.0005
0.0004	22.9017	1000	1054.5455	0.0004
0.0004	25.1618	1100	1045.4545	0.0003
0.0003	27.4393	1200	1234.0909	0.0003
0.0003	29.7168	1300	1247.7273	0.0002
0.0002	31.9942	1400	1377.2727	0.0002
0.0002	34.2543	1500	1631.8182	0.0002
0.0002	36.5318	1600	1625.0	0.0002
0.0002	38.8092	1700	1625.0	0.0002
0.0001	41.0694	1800	1595.4545	0.0001
0.0001	43.3468	1900	1618.1818	0.0001
0.0001	45.6243	2000	1602.2727	0.0001
0.0001	47.9017	2100	1600.0	0.0001
0.0001	50.1618	2200	1852.2727	0.0001
0.0001	52.4393	2300	1747.7273	0.0001
0.0001	54.7168	2400	1825.0	0.0001
0.0001	56.9942	2500	2131.8182	0.0001
0.0001	59.2543	2600	2270.4545	0.0001
0.0001	61.5318	2700	2127.2727	0.0001
0.0001	63.8092	2800	2365.9091	0.0001
0.0001	66.0694	2900	2363.6364	0.0001
0.0001	68.3468	3000	2490.9091	0.0001
0.0001	70.6243	3100	2488.6364	0.0001
0.0001	72.9017	3200	2488.6364	0.0001
0.0001	75.1618	3300	2488.6364	0.0001
0.0001	77.4393	3400	2488.6364	0.0001
0.0001	79.7168	3500	2488.6364	0.0001
0.0001	83.6936	3600	0.0001	2890.9091
0.0001	85.9711	3700	0.0001	2486.3636
0.0001	88.2312	3800	0.0001	2486.3636
0.0001	90.5087	3900	0.0001	2486.3636
0.0001	92.7861	4000	0.0001	2486.3636

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1